blob: 7096c22041e7e6680b320ca08e4be831591b6477 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Block device elevator/IO-scheduler.
3 *
4 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 *
Jens Axboe0fe23472006-09-04 15:41:16 +02006 * 30042000 Jens Axboe <axboe@kernel.dk> :
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 *
8 * Split the elevator a bit so that it is possible to choose a different
9 * one or even write a new "plug in". There are three pieces:
10 * - elevator_fn, inserts a new request in the queue list
11 * - elevator_merge_fn, decides whether a new buffer can be merged with
12 * an existing request
13 * - elevator_dequeue_fn, called when a request is taken off the active list
14 *
15 * 20082000 Dave Jones <davej@suse.de> :
16 * Removed tests for max-bomb-segments, which was breaking elvtune
17 * when run without -bN
18 *
19 * Jens:
20 * - Rework again to work with bio instead of buffer_heads
21 * - loose bi_dev comparisons, partition handling is right now
22 * - completely modularize elevator setup and teardown
23 *
24 */
25#include <linux/kernel.h>
26#include <linux/fs.h>
27#include <linux/blkdev.h>
28#include <linux/elevator.h>
29#include <linux/bio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/module.h>
31#include <linux/slab.h>
32#include <linux/init.h>
33#include <linux/compiler.h>
Jens Axboe2056a782006-03-23 20:00:26 +010034#include <linux/blktrace_api.h>
Jens Axboe98170642006-07-28 09:23:08 +020035#include <linux/hash.h>
Jens Axboe0835da62008-08-26 09:15:47 +020036#include <linux/uaccess.h>
Lin Mingc8158812013-03-23 11:42:27 +080037#include <linux/pm_runtime.h>
Tejun Heoeea8f412015-05-22 17:13:17 -040038#include <linux/blk-cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Li Zefan55782132009-06-09 13:43:05 +080040#include <trace/events/block.h>
41
Jens Axboe242f9dc2008-09-14 05:55:09 -070042#include "blk.h"
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044static DEFINE_SPINLOCK(elv_list_lock);
45static LIST_HEAD(elv_list);
46
47/*
Jens Axboe98170642006-07-28 09:23:08 +020048 * Merge hash stuff.
49 */
Tejun Heo83096eb2009-05-07 22:24:39 +090050#define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq))
Jens Axboe98170642006-07-28 09:23:08 +020051
52/*
Jens Axboeda775262006-12-20 11:04:12 +010053 * Query io scheduler to see if the current process issuing bio may be
54 * merged with rq.
55 */
Tahsin Erdogan72ef7992016-07-07 11:48:22 -070056static int elv_iosched_allow_bio_merge(struct request *rq, struct bio *bio)
Jens Axboeda775262006-12-20 11:04:12 +010057{
Jens Axboe165125e2007-07-24 09:28:11 +020058 struct request_queue *q = rq->q;
Jens Axboeb374d182008-10-31 10:05:07 +010059 struct elevator_queue *e = q->elevator;
Jens Axboeda775262006-12-20 11:04:12 +010060
Tahsin Erdogan72ef7992016-07-07 11:48:22 -070061 if (e->type->ops.elevator_allow_bio_merge_fn)
62 return e->type->ops.elevator_allow_bio_merge_fn(q, rq, bio);
Jens Axboeda775262006-12-20 11:04:12 +010063
64 return 1;
65}
66
67/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 * can we safely merge with this request?
69 */
Tahsin Erdogan72ef7992016-07-07 11:48:22 -070070bool elv_bio_merge_ok(struct request *rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071{
Tejun Heo050c8ea2012-02-08 09:19:38 +010072 if (!blk_rq_merge_ok(rq, bio))
Tahsin Erdogan72ef7992016-07-07 11:48:22 -070073 return false;
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +020074
Tahsin Erdogan72ef7992016-07-07 11:48:22 -070075 if (!elv_iosched_allow_bio_merge(rq, bio))
76 return false;
Jens Axboeda775262006-12-20 11:04:12 +010077
Tahsin Erdogan72ef7992016-07-07 11:48:22 -070078 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079}
Tahsin Erdogan72ef7992016-07-07 11:48:22 -070080EXPORT_SYMBOL(elv_bio_merge_ok);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Linus Torvalds1da177e2005-04-16 15:20:36 -070082static struct elevator_type *elevator_find(const char *name)
83{
Vasily Tarasova22b1692006-10-11 09:24:27 +020084 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Matthias Kaehlcke70cee262007-07-10 12:26:24 +020086 list_for_each_entry(e, &elv_list, list) {
Vasily Tarasova22b1692006-10-11 09:24:27 +020087 if (!strcmp(e->elevator_name, name))
88 return e;
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Vasily Tarasova22b1692006-10-11 09:24:27 +020091 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092}
93
94static void elevator_put(struct elevator_type *e)
95{
96 module_put(e->elevator_owner);
97}
98
Tejun Heo21c3c5d2013-01-22 16:48:03 -080099static struct elevator_type *elevator_get(const char *name, bool try_loading)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100{
Tejun Heo2824bc932005-10-20 10:56:41 +0200101 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200103 spin_lock(&elv_list_lock);
Tejun Heo2824bc932005-10-20 10:56:41 +0200104
105 e = elevator_find(name);
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800106 if (!e && try_loading) {
Jens Axboee1640942008-02-19 10:20:37 +0100107 spin_unlock(&elv_list_lock);
Kees Cook490b94b2011-05-05 18:02:12 -0600108 request_module("%s-iosched", name);
Jens Axboee1640942008-02-19 10:20:37 +0100109 spin_lock(&elv_list_lock);
110 e = elevator_find(name);
111 }
112
Tejun Heo2824bc932005-10-20 10:56:41 +0200113 if (e && !try_module_get(e->elevator_owner))
114 e = NULL;
115
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200116 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
118 return e;
119}
120
Wang Sheng-Hui484fc252011-09-08 12:32:14 +0200121static char chosen_elevator[ELV_NAME_MAX];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Nate Diller5f003972006-01-24 10:07:58 +0100123static int __init elevator_setup(char *str)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124{
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100125 /*
126 * Be backwards-compatible with previous kernels, so users
127 * won't get the wrong elevator.
128 */
Jens Axboe492af632009-10-03 09:37:51 +0200129 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800130 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131}
132
133__setup("elevator=", elevator_setup);
134
Tejun Heobb813f42013-01-18 14:05:56 -0800135/* called during boot to load the elevator chosen by the elevator param */
136void __init load_default_elevator_module(void)
137{
138 struct elevator_type *e;
139
140 if (!chosen_elevator[0])
141 return;
142
143 spin_lock(&elv_list_lock);
144 e = elevator_find(chosen_elevator);
145 spin_unlock(&elv_list_lock);
146
147 if (!e)
148 request_module("%s-iosched", chosen_elevator);
149}
150
Al Viro3d1ab402006-03-18 18:35:43 -0500151static struct kobj_type elv_ktype;
152
Jianpeng Mad50235b2013-07-03 13:25:24 +0200153struct elevator_queue *elevator_alloc(struct request_queue *q,
Jens Axboe165125e2007-07-24 09:28:11 +0200154 struct elevator_type *e)
Al Viro3d1ab402006-03-18 18:35:43 -0500155{
Jens Axboeb374d182008-10-31 10:05:07 +0100156 struct elevator_queue *eq;
Jens Axboe98170642006-07-28 09:23:08 +0200157
Joe Perchesc1b511e2013-08-29 15:21:42 -0700158 eq = kzalloc_node(sizeof(*eq), GFP_KERNEL, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200159 if (unlikely(!eq))
Chao Yu8406a4d2015-04-23 10:47:44 -0600160 return NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200161
Tejun Heo22f746e2011-12-14 00:33:41 +0100162 eq->type = e;
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -0700163 kobject_init(&eq->kobj, &elv_ktype);
Jens Axboe98170642006-07-28 09:23:08 +0200164 mutex_init(&eq->sysfs_lock);
Sasha Levin242d98f2012-12-17 10:01:27 -0500165 hash_init(eq->hash);
Jens Axboe98170642006-07-28 09:23:08 +0200166
Al Viro3d1ab402006-03-18 18:35:43 -0500167 return eq;
168}
Jianpeng Mad50235b2013-07-03 13:25:24 +0200169EXPORT_SYMBOL(elevator_alloc);
Al Viro3d1ab402006-03-18 18:35:43 -0500170
171static void elevator_release(struct kobject *kobj)
172{
Jens Axboeb374d182008-10-31 10:05:07 +0100173 struct elevator_queue *e;
Jens Axboe98170642006-07-28 09:23:08 +0200174
Jens Axboeb374d182008-10-31 10:05:07 +0100175 e = container_of(kobj, struct elevator_queue, kobj);
Tejun Heo22f746e2011-12-14 00:33:41 +0100176 elevator_put(e->type);
Al Viro3d1ab402006-03-18 18:35:43 -0500177 kfree(e);
178}
179
Jens Axboe165125e2007-07-24 09:28:11 +0200180int elevator_init(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181{
182 struct elevator_type *e = NULL;
Tejun Heof8fc8772011-12-14 00:33:40 +0100183 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600185 /*
186 * q->sysfs_lock must be held to provide mutual exclusion between
187 * elevator_switch() and here.
188 */
189 lockdep_assert_held(&q->sysfs_lock);
190
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400191 if (unlikely(q->elevator))
192 return 0;
193
Tejun Heocb98fc82005-10-28 08:29:39 +0200194 INIT_LIST_HEAD(&q->queue_head);
195 q->last_merge = NULL;
196 q->end_sector = 0;
197 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200198
Jens Axboe4eb166d2008-02-01 00:37:27 +0100199 if (name) {
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800200 e = elevator_get(name, true);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100201 if (!e)
202 return -EINVAL;
203 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800205 /*
206 * Use the default elevator specified by config boot param or
207 * config option. Don't try to load modules as we could be running
208 * off async and request_module() isn't allowed from async.
209 */
Jens Axboe4eb166d2008-02-01 00:37:27 +0100210 if (!e && *chosen_elevator) {
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800211 e = elevator_get(chosen_elevator, false);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100212 if (!e)
213 printk(KERN_ERR "I/O scheduler %s not found\n",
214 chosen_elevator);
215 }
Nate Diller248d5ca2006-01-24 10:09:14 +0100216
Jens Axboe4eb166d2008-02-01 00:37:27 +0100217 if (!e) {
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800218 e = elevator_get(CONFIG_DEFAULT_IOSCHED, false);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100219 if (!e) {
220 printk(KERN_ERR
221 "Default I/O scheduler not found. " \
222 "Using noop.\n");
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800223 e = elevator_get("noop", false);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100224 }
Nate Diller5f003972006-01-24 10:07:58 +0100225 }
226
Jianpeng Mad50235b2013-07-03 13:25:24 +0200227 err = e->ops.elevator_init_fn(q, e);
Sudip Mukherjeed32f6b52014-10-23 22:16:48 +0530228 if (err)
229 elevator_put(e);
230 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231}
Jens Axboe2e662b62006-07-13 11:55:04 +0200232EXPORT_SYMBOL(elevator_init);
233
Jens Axboeb374d182008-10-31 10:05:07 +0100234void elevator_exit(struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235{
Al Viro3d1ab402006-03-18 18:35:43 -0500236 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100237 if (e->type->ops.elevator_exit_fn)
238 e->type->ops.elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500239 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
Al Viro3d1ab402006-03-18 18:35:43 -0500241 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242}
Jens Axboe2e662b62006-07-13 11:55:04 +0200243EXPORT_SYMBOL(elevator_exit);
244
Jens Axboe98170642006-07-28 09:23:08 +0200245static inline void __elv_rqhash_del(struct request *rq)
246{
Sasha Levin242d98f2012-12-17 10:01:27 -0500247 hash_del(&rq->hash);
Jens Axboe360f92c2014-04-09 20:27:01 -0600248 rq->cmd_flags &= ~REQ_HASHED;
Jens Axboe98170642006-07-28 09:23:08 +0200249}
250
Jens Axboe165125e2007-07-24 09:28:11 +0200251static void elv_rqhash_del(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200252{
253 if (ELV_ON_HASH(rq))
254 __elv_rqhash_del(rq);
255}
256
Jens Axboe165125e2007-07-24 09:28:11 +0200257static void elv_rqhash_add(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200258{
Jens Axboeb374d182008-10-31 10:05:07 +0100259 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200260
261 BUG_ON(ELV_ON_HASH(rq));
Sasha Levin242d98f2012-12-17 10:01:27 -0500262 hash_add(e->hash, &rq->hash, rq_hash_key(rq));
Jens Axboe360f92c2014-04-09 20:27:01 -0600263 rq->cmd_flags |= REQ_HASHED;
Jens Axboe98170642006-07-28 09:23:08 +0200264}
265
Jens Axboe165125e2007-07-24 09:28:11 +0200266static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200267{
268 __elv_rqhash_del(rq);
269 elv_rqhash_add(q, rq);
270}
271
Jens Axboe165125e2007-07-24 09:28:11 +0200272static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
Jens Axboe98170642006-07-28 09:23:08 +0200273{
Jens Axboeb374d182008-10-31 10:05:07 +0100274 struct elevator_queue *e = q->elevator;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800275 struct hlist_node *next;
Jens Axboe98170642006-07-28 09:23:08 +0200276 struct request *rq;
277
Linus Torvaldsee89f812013-02-28 12:52:24 -0800278 hash_for_each_possible_safe(e->hash, rq, next, hash, offset) {
Jens Axboe98170642006-07-28 09:23:08 +0200279 BUG_ON(!ELV_ON_HASH(rq));
280
281 if (unlikely(!rq_mergeable(rq))) {
282 __elv_rqhash_del(rq);
283 continue;
284 }
285
286 if (rq_hash_key(rq) == offset)
287 return rq;
288 }
289
290 return NULL;
291}
292
Tejun Heo8922e162005-10-20 16:23:44 +0200293/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200294 * RB-tree support functions for inserting/lookup/removal of requests
295 * in a sorted RB tree.
296 */
Jeff Moyer796d5112011-06-02 21:19:05 +0200297void elv_rb_add(struct rb_root *root, struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +0200298{
299 struct rb_node **p = &root->rb_node;
300 struct rb_node *parent = NULL;
301 struct request *__rq;
302
303 while (*p) {
304 parent = *p;
305 __rq = rb_entry(parent, struct request, rb_node);
306
Tejun Heo83096eb2009-05-07 22:24:39 +0900307 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200308 p = &(*p)->rb_left;
Jeff Moyer796d5112011-06-02 21:19:05 +0200309 else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200310 p = &(*p)->rb_right;
Jens Axboe2e662b62006-07-13 11:55:04 +0200311 }
312
313 rb_link_node(&rq->rb_node, parent, p);
314 rb_insert_color(&rq->rb_node, root);
Jens Axboe2e662b62006-07-13 11:55:04 +0200315}
Jens Axboe2e662b62006-07-13 11:55:04 +0200316EXPORT_SYMBOL(elv_rb_add);
317
318void elv_rb_del(struct rb_root *root, struct request *rq)
319{
320 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
321 rb_erase(&rq->rb_node, root);
322 RB_CLEAR_NODE(&rq->rb_node);
323}
Jens Axboe2e662b62006-07-13 11:55:04 +0200324EXPORT_SYMBOL(elv_rb_del);
325
326struct request *elv_rb_find(struct rb_root *root, sector_t sector)
327{
328 struct rb_node *n = root->rb_node;
329 struct request *rq;
330
331 while (n) {
332 rq = rb_entry(n, struct request, rb_node);
333
Tejun Heo83096eb2009-05-07 22:24:39 +0900334 if (sector < blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200335 n = n->rb_left;
Tejun Heo83096eb2009-05-07 22:24:39 +0900336 else if (sector > blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200337 n = n->rb_right;
338 else
339 return rq;
340 }
341
342 return NULL;
343}
Jens Axboe2e662b62006-07-13 11:55:04 +0200344EXPORT_SYMBOL(elv_rb_find);
345
346/*
Tejun Heo8922e162005-10-20 16:23:44 +0200347 * Insert rq into dispatch queue of q. Queue lock must be held on
Uwe Kleine-Königdbe7f762007-10-20 01:55:04 +0200348 * entry. rq is sort instead into the dispatch queue. To be used by
Jens Axboe2e662b62006-07-13 11:55:04 +0200349 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200350 */
Jens Axboe165125e2007-07-24 09:28:11 +0200351void elv_dispatch_sort(struct request_queue *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200352{
353 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200354 struct list_head *entry;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100355 int stop_flags;
Tejun Heo8922e162005-10-20 16:23:44 +0200356
Tejun Heo06b86242005-10-20 16:46:23 +0200357 if (q->last_merge == rq)
358 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200359
360 elv_rqhash_del(q, rq);
361
Tejun Heo15853af2005-11-10 08:52:05 +0100362 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200363
Jens Axboe1b47f532005-10-20 16:37:00 +0200364 boundary = q->end_sector;
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100365 stop_flags = REQ_SOFTBARRIER | REQ_STARTED;
Tejun Heo8922e162005-10-20 16:23:44 +0200366 list_for_each_prev(entry, &q->queue_head) {
367 struct request *pos = list_entry_rq(entry);
368
Mike Christieba568ea2016-06-05 14:32:13 -0500369 if ((req_op(rq) == REQ_OP_DISCARD) != (req_op(pos) == REQ_OP_DISCARD))
David Woodhousee17fc0a2008-08-09 16:42:20 +0100370 break;
Jens Axboe783660b2007-01-19 11:27:47 +1100371 if (rq_data_dir(rq) != rq_data_dir(pos))
372 break;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100373 if (pos->cmd_flags & stop_flags)
Tejun Heo8922e162005-10-20 16:23:44 +0200374 break;
Tejun Heo83096eb2009-05-07 22:24:39 +0900375 if (blk_rq_pos(rq) >= boundary) {
376 if (blk_rq_pos(pos) < boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200377 continue;
378 } else {
Tejun Heo83096eb2009-05-07 22:24:39 +0900379 if (blk_rq_pos(pos) >= boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200380 break;
381 }
Tejun Heo83096eb2009-05-07 22:24:39 +0900382 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
Tejun Heo8922e162005-10-20 16:23:44 +0200383 break;
384 }
385
386 list_add(&rq->queuelist, entry);
387}
Jens Axboe2e662b62006-07-13 11:55:04 +0200388EXPORT_SYMBOL(elv_dispatch_sort);
389
Jens Axboe98170642006-07-28 09:23:08 +0200390/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200391 * Insert rq into dispatch queue of q. Queue lock must be held on
392 * entry. rq is added to the back of the dispatch queue. To be used by
393 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200394 */
395void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
396{
397 if (q->last_merge == rq)
398 q->last_merge = NULL;
399
400 elv_rqhash_del(q, rq);
401
402 q->nr_sorted--;
403
404 q->end_sector = rq_end_sector(rq);
405 q->boundary_rq = rq;
406 list_add_tail(&rq->queuelist, &q->queue_head);
407}
Jens Axboe2e662b62006-07-13 11:55:04 +0200408EXPORT_SYMBOL(elv_dispatch_add_tail);
409
Jens Axboe165125e2007-07-24 09:28:11 +0200410int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411{
Jens Axboeb374d182008-10-31 10:05:07 +0100412 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200413 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200414 int ret;
415
Jens Axboe98170642006-07-28 09:23:08 +0200416 /*
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100417 * Levels of merges:
418 * nomerges: No merges at all attempted
419 * noxmerges: Only simple one-hit cache try
420 * merges: All merge tries attempted
421 */
Ming Lei7460d382015-10-20 23:13:55 +0800422 if (blk_queue_nomerges(q) || !bio_mergeable(bio))
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100423 return ELEVATOR_NO_MERGE;
424
425 /*
Jens Axboe98170642006-07-28 09:23:08 +0200426 * First try one-hit cache.
427 */
Tahsin Erdogan72ef7992016-07-07 11:48:22 -0700428 if (q->last_merge && elv_bio_merge_ok(q->last_merge, bio)) {
Tejun Heo050c8ea2012-02-08 09:19:38 +0100429 ret = blk_try_merge(q->last_merge, bio);
Tejun Heo06b86242005-10-20 16:46:23 +0200430 if (ret != ELEVATOR_NO_MERGE) {
431 *req = q->last_merge;
432 return ret;
433 }
434 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100436 if (blk_queue_noxmerges(q))
Alan D. Brunelleac9fafa2008-04-29 14:44:19 +0200437 return ELEVATOR_NO_MERGE;
438
Jens Axboe98170642006-07-28 09:23:08 +0200439 /*
440 * See if our hash lookup can find a potential backmerge.
441 */
Kent Overstreet4f024f32013-10-11 15:44:27 -0700442 __rq = elv_rqhash_find(q, bio->bi_iter.bi_sector);
Tahsin Erdogan72ef7992016-07-07 11:48:22 -0700443 if (__rq && elv_bio_merge_ok(__rq, bio)) {
Jens Axboe98170642006-07-28 09:23:08 +0200444 *req = __rq;
445 return ELEVATOR_BACK_MERGE;
446 }
447
Tejun Heo22f746e2011-12-14 00:33:41 +0100448 if (e->type->ops.elevator_merge_fn)
449 return e->type->ops.elevator_merge_fn(q, req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
451 return ELEVATOR_NO_MERGE;
452}
453
Jens Axboe5e84ea32011-03-21 10:14:27 +0100454/*
455 * Attempt to do an insertion back merge. Only check for the case where
456 * we can append 'rq' to an existing request, so we can throw 'rq' away
457 * afterwards.
458 *
459 * Returns true if we merged, false otherwise
460 */
461static bool elv_attempt_insert_merge(struct request_queue *q,
462 struct request *rq)
463{
464 struct request *__rq;
Shaohua Libee03932012-11-09 08:44:27 +0100465 bool ret;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100466
467 if (blk_queue_nomerges(q))
468 return false;
469
470 /*
471 * First try one-hit cache.
472 */
473 if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
474 return true;
475
476 if (blk_queue_noxmerges(q))
477 return false;
478
Shaohua Libee03932012-11-09 08:44:27 +0100479 ret = false;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100480 /*
481 * See if our hash lookup can find a potential backmerge.
482 */
Shaohua Libee03932012-11-09 08:44:27 +0100483 while (1) {
484 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
485 if (!__rq || !blk_attempt_req_merge(q, __rq, rq))
486 break;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100487
Shaohua Libee03932012-11-09 08:44:27 +0100488 /* The merged request could be merged with others, try again */
489 ret = true;
490 rq = __rq;
491 }
492
493 return ret;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100494}
495
Jens Axboe165125e2007-07-24 09:28:11 +0200496void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497{
Jens Axboeb374d182008-10-31 10:05:07 +0100498 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
Tejun Heo22f746e2011-12-14 00:33:41 +0100500 if (e->type->ops.elevator_merged_fn)
501 e->type->ops.elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200502
Jens Axboe2e662b62006-07-13 11:55:04 +0200503 if (type == ELEVATOR_BACK_MERGE)
504 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200505
Tejun Heo06b86242005-10-20 16:46:23 +0200506 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507}
508
Jens Axboe165125e2007-07-24 09:28:11 +0200509void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 struct request *next)
511{
Jens Axboeb374d182008-10-31 10:05:07 +0100512 struct elevator_queue *e = q->elevator;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100513 const int next_sorted = next->cmd_flags & REQ_SORTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
Tejun Heo22f746e2011-12-14 00:33:41 +0100515 if (next_sorted && e->type->ops.elevator_merge_req_fn)
516 e->type->ops.elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200517
Jens Axboe98170642006-07-28 09:23:08 +0200518 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200519
Jens Axboe5e84ea32011-03-21 10:14:27 +0100520 if (next_sorted) {
521 elv_rqhash_del(q, next);
522 q->nr_sorted--;
523 }
524
Tejun Heo06b86242005-10-20 16:46:23 +0200525 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526}
527
Divyesh Shah812d4022010-04-08 21:14:23 -0700528void elv_bio_merged(struct request_queue *q, struct request *rq,
529 struct bio *bio)
530{
531 struct elevator_queue *e = q->elevator;
532
Tejun Heo22f746e2011-12-14 00:33:41 +0100533 if (e->type->ops.elevator_bio_merged_fn)
534 e->type->ops.elevator_bio_merged_fn(q, rq, bio);
Divyesh Shah812d4022010-04-08 21:14:23 -0700535}
536
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +0100537#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +0800538static void blk_pm_requeue_request(struct request *rq)
539{
540 if (rq->q->dev && !(rq->cmd_flags & REQ_PM))
541 rq->q->nr_pending--;
542}
543
544static void blk_pm_add_request(struct request_queue *q, struct request *rq)
545{
546 if (q->dev && !(rq->cmd_flags & REQ_PM) && q->nr_pending++ == 0 &&
547 (q->rpm_status == RPM_SUSPENDED || q->rpm_status == RPM_SUSPENDING))
548 pm_request_resume(q->dev);
549}
550#else
551static inline void blk_pm_requeue_request(struct request *rq) {}
552static inline void blk_pm_add_request(struct request_queue *q,
553 struct request *rq)
554{
555}
556#endif
557
Jens Axboe165125e2007-07-24 09:28:11 +0200558void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 /*
561 * it already went through dequeue, we need to decrement the
562 * in_flight count again
563 */
Tejun Heo8922e162005-10-20 16:23:44 +0200564 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200565 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200566 if (rq->cmd_flags & REQ_SORTED)
Jens Axboecad97512007-01-14 22:26:09 +1100567 elv_deactivate_rq(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200568 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Jens Axboe4aff5e22006-08-10 08:44:47 +0200570 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
Lin Mingc8158812013-03-23 11:42:27 +0800572 blk_pm_requeue_request(rq);
573
Jens Axboeb710a482011-03-30 09:52:30 +0200574 __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575}
576
Jerome Marchand26308ea2009-03-27 10:31:51 +0100577void elv_drain_elevator(struct request_queue *q)
Tejun Heo15853af2005-11-10 08:52:05 +0100578{
579 static int printed;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200580
581 lockdep_assert_held(q->queue_lock);
582
Tejun Heo22f746e2011-12-14 00:33:41 +0100583 while (q->elevator->type->ops.elevator_dispatch_fn(q, 1))
Tejun Heo15853af2005-11-10 08:52:05 +0100584 ;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200585 if (q->nr_sorted && printed++ < 10) {
Tejun Heo15853af2005-11-10 08:52:05 +0100586 printk(KERN_ERR "%s: forced dispatching is broken "
587 "(nr_sorted=%u), please report this\n",
Tejun Heo22f746e2011-12-14 00:33:41 +0100588 q->elevator->type->elevator_name, q->nr_sorted);
Tejun Heo15853af2005-11-10 08:52:05 +0100589 }
590}
591
Jens Axboeb710a482011-03-30 09:52:30 +0200592void __elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593{
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100594 trace_block_rq_insert(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100595
Lin Mingc8158812013-03-23 11:42:27 +0800596 blk_pm_add_request(q, rq);
597
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 rq->q = q;
599
Jens Axboeb710a482011-03-30 09:52:30 +0200600 if (rq->cmd_flags & REQ_SOFTBARRIER) {
601 /* barriers are scheduling boundary, update end_sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400602 if (rq->cmd_type == REQ_TYPE_FS) {
Jens Axboeb710a482011-03-30 09:52:30 +0200603 q->end_sector = rq_end_sector(rq);
604 q->boundary_rq = rq;
605 }
606 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
Jens Axboe3aa72872011-04-21 19:28:35 +0200607 (where == ELEVATOR_INSERT_SORT ||
608 where == ELEVATOR_INSERT_SORT_MERGE))
Jens Axboeb710a482011-03-30 09:52:30 +0200609 where = ELEVATOR_INSERT_BACK;
610
Tejun Heo8922e162005-10-20 16:23:44 +0200611 switch (where) {
Tejun Heo28e7d182010-09-03 11:56:16 +0200612 case ELEVATOR_INSERT_REQUEUE:
Tejun Heo8922e162005-10-20 16:23:44 +0200613 case ELEVATOR_INSERT_FRONT:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200614 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200615 list_add(&rq->queuelist, &q->queue_head);
616 break;
617
618 case ELEVATOR_INSERT_BACK:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200619 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100620 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200621 list_add_tail(&rq->queuelist, &q->queue_head);
622 /*
623 * We kick the queue here for the following reasons.
624 * - The elevator might have returned NULL previously
625 * to delay requests and returned them now. As the
626 * queue wasn't empty before this request, ll_rw_blk
627 * won't run the queue on return, resulting in hang.
628 * - Usually, back inserted requests won't be merged
629 * with anything. There's no point in delaying queue
630 * processing.
631 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200632 __blk_run_queue(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200633 break;
634
Jens Axboe5e84ea32011-03-21 10:14:27 +0100635 case ELEVATOR_INSERT_SORT_MERGE:
636 /*
637 * If we succeed in merging this request with one in the
638 * queue already, we are done - rq has now been freed,
639 * so no need to do anything further.
640 */
641 if (elv_attempt_insert_merge(q, rq))
642 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200643 case ELEVATOR_INSERT_SORT:
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400644 BUG_ON(rq->cmd_type != REQ_TYPE_FS);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200645 rq->cmd_flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100646 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200647 if (rq_mergeable(rq)) {
648 elv_rqhash_add(q, rq);
649 if (!q->last_merge)
650 q->last_merge = rq;
651 }
652
Tejun Heoca235092005-11-01 17:23:49 +0900653 /*
654 * Some ioscheds (cfq) run q->request_fn directly, so
655 * rq cannot be accessed after calling
656 * elevator_add_req_fn.
657 */
Tejun Heo22f746e2011-12-14 00:33:41 +0100658 q->elevator->type->ops.elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200659 break;
660
Tejun Heoae1b1532011-01-25 12:43:54 +0100661 case ELEVATOR_INSERT_FLUSH:
662 rq->cmd_flags |= REQ_SOFTBARRIER;
663 blk_insert_flush(rq);
664 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200665 default:
666 printk(KERN_ERR "%s: bad insertion point %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700667 __func__, where);
Tejun Heo8922e162005-10-20 16:23:44 +0200668 BUG();
669 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670}
Jens Axboe2e662b62006-07-13 11:55:04 +0200671EXPORT_SYMBOL(__elv_add_request);
672
Jens Axboe7eaceac2011-03-10 08:52:07 +0100673void elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674{
675 unsigned long flags;
676
677 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100678 __elv_add_request(q, rq, where);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 spin_unlock_irqrestore(q->queue_lock, flags);
680}
Jens Axboe2e662b62006-07-13 11:55:04 +0200681EXPORT_SYMBOL(elv_add_request);
682
Jens Axboe165125e2007-07-24 09:28:11 +0200683struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684{
Jens Axboeb374d182008-10-31 10:05:07 +0100685 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
Tejun Heo22f746e2011-12-14 00:33:41 +0100687 if (e->type->ops.elevator_latter_req_fn)
688 return e->type->ops.elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 return NULL;
690}
691
Jens Axboe165125e2007-07-24 09:28:11 +0200692struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693{
Jens Axboeb374d182008-10-31 10:05:07 +0100694 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
Tejun Heo22f746e2011-12-14 00:33:41 +0100696 if (e->type->ops.elevator_former_req_fn)
697 return e->type->ops.elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 return NULL;
699}
700
Tejun Heo852c7882012-03-05 13:15:27 -0800701int elv_set_request(struct request_queue *q, struct request *rq,
702 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703{
Jens Axboeb374d182008-10-31 10:05:07 +0100704 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
Tejun Heo22f746e2011-12-14 00:33:41 +0100706 if (e->type->ops.elevator_set_req_fn)
Tejun Heo852c7882012-03-05 13:15:27 -0800707 return e->type->ops.elevator_set_req_fn(q, rq, bio, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 return 0;
709}
710
Jens Axboe165125e2007-07-24 09:28:11 +0200711void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712{
Jens Axboeb374d182008-10-31 10:05:07 +0100713 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
Tejun Heo22f746e2011-12-14 00:33:41 +0100715 if (e->type->ops.elevator_put_req_fn)
716 e->type->ops.elevator_put_req_fn(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
718
Mike Christieba568ea2016-06-05 14:32:13 -0500719int elv_may_queue(struct request_queue *q, int op, int op_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720{
Jens Axboeb374d182008-10-31 10:05:07 +0100721 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
Tejun Heo22f746e2011-12-14 00:33:41 +0100723 if (e->type->ops.elevator_may_queue_fn)
Mike Christieba568ea2016-06-05 14:32:13 -0500724 return e->type->ops.elevator_may_queue_fn(q, op, op_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
726 return ELV_MQUEUE_MAY;
727}
728
Jens Axboe165125e2007-07-24 09:28:11 +0200729void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730{
Jens Axboeb374d182008-10-31 10:05:07 +0100731 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
733 /*
734 * request is released from the driver, io must be done
735 */
Tejun Heo8922e162005-10-20 16:23:44 +0200736 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200737 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200738 if ((rq->cmd_flags & REQ_SORTED) &&
Tejun Heo22f746e2011-12-14 00:33:41 +0100739 e->type->ops.elevator_completed_req_fn)
740 e->type->ops.elevator_completed_req_fn(q, rq);
Tejun Heo1bc691d2006-01-12 15:39:26 +0100741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742}
743
Al Viro3d1ab402006-03-18 18:35:43 -0500744#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
745
746static ssize_t
747elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
748{
Al Viro3d1ab402006-03-18 18:35:43 -0500749 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100750 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500751 ssize_t error;
752
753 if (!entry->show)
754 return -EIO;
755
Jens Axboeb374d182008-10-31 10:05:07 +0100756 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500757 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100758 error = e->type ? entry->show(e, page) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500759 mutex_unlock(&e->sysfs_lock);
760 return error;
761}
762
763static ssize_t
764elv_attr_store(struct kobject *kobj, struct attribute *attr,
765 const char *page, size_t length)
766{
Al Viro3d1ab402006-03-18 18:35:43 -0500767 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100768 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500769 ssize_t error;
770
771 if (!entry->store)
772 return -EIO;
773
Jens Axboeb374d182008-10-31 10:05:07 +0100774 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500775 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100776 error = e->type ? entry->store(e, page, length) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500777 mutex_unlock(&e->sysfs_lock);
778 return error;
779}
780
Emese Revfy52cf25d2010-01-19 02:58:23 +0100781static const struct sysfs_ops elv_sysfs_ops = {
Al Viro3d1ab402006-03-18 18:35:43 -0500782 .show = elv_attr_show,
783 .store = elv_attr_store,
784};
785
786static struct kobj_type elv_ktype = {
787 .sysfs_ops = &elv_sysfs_ops,
788 .release = elevator_release,
789};
790
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800791int elv_register_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792{
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800793 struct elevator_queue *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500794 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -0700796 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
Al Viro3d1ab402006-03-18 18:35:43 -0500797 if (!error) {
Tejun Heo22f746e2011-12-14 00:33:41 +0100798 struct elv_fs_entry *attr = e->type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500799 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500800 while (attr->attr.name) {
801 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500802 break;
Al Viroe572ec72006-03-18 22:27:18 -0500803 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500804 }
805 }
806 kobject_uevent(&e->kobj, KOBJ_ADD);
Jens Axboe430c62f2010-10-07 09:35:16 +0200807 e->registered = 1;
Jens Axboe0bb97942015-06-10 08:01:20 -0600808 if (e->type->ops.elevator_registered_fn)
809 e->type->ops.elevator_registered_fn(q);
Al Viro3d1ab402006-03-18 18:35:43 -0500810 }
811 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812}
Tejun Heof8fc8772011-12-14 00:33:40 +0100813EXPORT_SYMBOL(elv_register_queue);
Jens Axboebc1c1162006-06-08 08:49:06 +0200814
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815void elv_unregister_queue(struct request_queue *q)
816{
Tejun Heof8fc8772011-12-14 00:33:40 +0100817 if (q) {
818 struct elevator_queue *e = q->elevator;
819
820 kobject_uevent(&e->kobj, KOBJ_REMOVE);
821 kobject_del(&e->kobj);
822 e->registered = 0;
823 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824}
Mike Snitzer01effb02010-05-11 08:57:42 +0200825EXPORT_SYMBOL(elv_unregister_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
Jens Axboee567bf72014-06-22 16:32:48 -0600827int elv_register(struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828{
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100829 char *def = "";
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200830
Tejun Heo3d3c2372011-12-14 00:33:42 +0100831 /* create icq_cache if requested */
832 if (e->icq_size) {
833 if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
834 WARN_ON(e->icq_align < __alignof__(struct io_cq)))
835 return -EINVAL;
836
837 snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
838 "%s_io_cq", e->elevator_name);
839 e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
840 e->icq_align, 0, NULL);
841 if (!e->icq_cache)
842 return -ENOMEM;
843 }
844
845 /* register, don't allow duplicate names */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200846 spin_lock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100847 if (elevator_find(e->elevator_name)) {
848 spin_unlock(&elv_list_lock);
849 if (e->icq_cache)
850 kmem_cache_destroy(e->icq_cache);
851 return -EBUSY;
852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 list_add_tail(&e->list, &elv_list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200854 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
Tejun Heo3d3c2372011-12-14 00:33:42 +0100856 /* print pretty message */
Nate Diller5f003972006-01-24 10:07:58 +0100857 if (!strcmp(e->elevator_name, chosen_elevator) ||
858 (!*chosen_elevator &&
859 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100860 def = " (default)";
861
Jens Axboe4eb166d2008-02-01 00:37:27 +0100862 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
863 def);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100864 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865}
866EXPORT_SYMBOL_GPL(elv_register);
867
868void elv_unregister(struct elevator_type *e)
869{
Tejun Heo3d3c2372011-12-14 00:33:42 +0100870 /* unregister */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200871 spin_lock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 list_del_init(&e->list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200873 spin_unlock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100874
875 /*
876 * Destroy icq_cache if it exists. icq's are RCU managed. Make
877 * sure all RCU operations are complete before proceeding.
878 */
879 if (e->icq_cache) {
880 rcu_barrier();
881 kmem_cache_destroy(e->icq_cache);
882 e->icq_cache = NULL;
883 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884}
885EXPORT_SYMBOL_GPL(elv_unregister);
886
887/*
888 * switch to new_e io scheduler. be careful not to introduce deadlocks -
889 * we don't free the old io scheduler, before we have allocated what we
890 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200891 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 */
Jens Axboe165125e2007-07-24 09:28:11 +0200893static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894{
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800895 struct elevator_queue *old = q->elevator;
896 bool registered = old->registered;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800897 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800899 /*
900 * Turn on BYPASS and drain all requests w/ elevator private data.
901 * Block layer doesn't call into a quiesced elevator - all requests
902 * are directly put on the dispatch list without elevator data
903 * using INSERT_BACK. All requests have SOFTBARRIER set and no
904 * merge happens either.
905 */
Tejun Heod7325802012-03-05 13:14:58 -0800906 blk_queue_bypass_start(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200907
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800908 /* unregister and clear all auxiliary data of the old elevator */
909 if (registered)
Tejun Heof8fc8772011-12-14 00:33:40 +0100910 elv_unregister_queue(q);
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800911
912 spin_lock_irq(q->queue_lock);
913 ioc_clear_queue(q);
914 spin_unlock_irq(q->queue_lock);
915
916 /* allocate, init and register new elevator */
Jianpeng Mad50235b2013-07-03 13:25:24 +0200917 err = new_e->ops.elevator_init_fn(q, new_e);
918 if (err)
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800919 goto fail_init;
920
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800921 if (registered) {
922 err = elv_register_queue(q);
Jens Axboe430c62f2010-10-07 09:35:16 +0200923 if (err)
924 goto fail_register;
925 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800927 /* done, kill the old one and finish */
928 elevator_exit(old);
Tejun Heod7325802012-03-05 13:14:58 -0800929 blk_queue_bypass_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200930
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800931 blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name);
Alan D. Brunelle4722dc52008-05-27 14:55:00 +0200932
Jens Axboe5dd531a2010-08-23 13:52:19 +0200933 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
935fail_register:
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800936 elevator_exit(q->elevator);
937fail_init:
938 /* switch failed, restore and re-register old elevator */
939 q->elevator = old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 elv_register_queue(q);
Tejun Heod7325802012-03-05 13:14:58 -0800941 blk_queue_bypass_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200942
Jens Axboe5dd531a2010-08-23 13:52:19 +0200943 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944}
945
Jens Axboe5dd531a2010-08-23 13:52:19 +0200946/*
947 * Switch this queue to the given IO scheduler.
948 */
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -0600949static int __elevator_change(struct request_queue *q, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950{
951 char elevator_name[ELV_NAME_MAX];
952 struct elevator_type *e;
953
Martin K. Petersencd43e262009-05-22 17:17:52 -0400954 if (!q->elevator)
Jens Axboe5dd531a2010-08-23 13:52:19 +0200955 return -ENXIO;
Martin K. Petersencd43e262009-05-22 17:17:52 -0400956
Li Zefanee2e9922008-10-14 08:49:56 +0200957 strlcpy(elevator_name, name, sizeof(elevator_name));
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800958 e = elevator_get(strstrip(elevator_name), true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 if (!e) {
960 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
961 return -EINVAL;
962 }
963
Tejun Heo22f746e2011-12-14 00:33:41 +0100964 if (!strcmp(elevator_name, q->elevator->type->elevator_name)) {
Nate Diller2ca7d932005-10-30 15:02:24 -0800965 elevator_put(e);
Jens Axboe5dd531a2010-08-23 13:52:19 +0200966 return 0;
Nate Diller2ca7d932005-10-30 15:02:24 -0800967 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
Jens Axboe5dd531a2010-08-23 13:52:19 +0200969 return elevator_switch(q, e);
970}
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -0600971
972int elevator_change(struct request_queue *q, const char *name)
973{
974 int ret;
975
976 /* Protect q->elevator from elevator_init() */
977 mutex_lock(&q->sysfs_lock);
978 ret = __elevator_change(q, name);
979 mutex_unlock(&q->sysfs_lock);
980
981 return ret;
982}
Jens Axboe5dd531a2010-08-23 13:52:19 +0200983EXPORT_SYMBOL(elevator_change);
984
985ssize_t elv_iosched_store(struct request_queue *q, const char *name,
986 size_t count)
987{
988 int ret;
989
990 if (!q->elevator)
991 return count;
992
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -0600993 ret = __elevator_change(q, name);
Jens Axboe5dd531a2010-08-23 13:52:19 +0200994 if (!ret)
995 return count;
996
997 printk(KERN_ERR "elevator: switch to %s failed\n", name);
998 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999}
1000
Jens Axboe165125e2007-07-24 09:28:11 +02001001ssize_t elv_iosched_show(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002{
Jens Axboeb374d182008-10-31 10:05:07 +01001003 struct elevator_queue *e = q->elevator;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001004 struct elevator_type *elv;
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001005 struct elevator_type *__e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 int len = 0;
1007
Mike Snitzere36f7242010-05-24 09:07:32 +02001008 if (!q->elevator || !blk_queue_stackable(q))
Martin K. Petersencd43e262009-05-22 17:17:52 -04001009 return sprintf(name, "none\n");
1010
Tejun Heo22f746e2011-12-14 00:33:41 +01001011 elv = e->type;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001012
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001013 spin_lock(&elv_list_lock);
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001014 list_for_each_entry(__e, &elv_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 if (!strcmp(elv->elevator_name, __e->elevator_name))
1016 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1017 else
1018 len += sprintf(name+len, "%s ", __e->elevator_name);
1019 }
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001020 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
1022 len += sprintf(len+name, "\n");
1023 return len;
1024}
1025
Jens Axboe165125e2007-07-24 09:28:11 +02001026struct request *elv_rb_former_request(struct request_queue *q,
1027 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001028{
1029 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1030
1031 if (rbprev)
1032 return rb_entry_rq(rbprev);
1033
1034 return NULL;
1035}
Jens Axboe2e662b62006-07-13 11:55:04 +02001036EXPORT_SYMBOL(elv_rb_former_request);
1037
Jens Axboe165125e2007-07-24 09:28:11 +02001038struct request *elv_rb_latter_request(struct request_queue *q,
1039 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001040{
1041 struct rb_node *rbnext = rb_next(&rq->rb_node);
1042
1043 if (rbnext)
1044 return rb_entry_rq(rbnext);
1045
1046 return NULL;
1047}
Jens Axboe2e662b62006-07-13 11:55:04 +02001048EXPORT_SYMBOL(elv_rb_latter_request);